dc.description.abstract | In this dissertation, two biometric-based pattern recognition problems were studied, i.e., off-line handwritten signature verification and human face recognition. Biometrics, by definition, is the automated technique of measuring a physical characteristic or person trait of an individual and comparing the characteristic or trait to a database for purposes of recognizing or authenticating that individual. Biometrics uses physical characteristics, defined as the things we are, and personal traits, defined as the things we behave, including facial thermographs, chemical composition of body odor, retina and iris, fingerprints, hand geometry, skin pores, wrist/hand veins, handwritten signature, keystrokes or typing, and voiceprint.
To deal with the first biometric-based pattern recognition problem, i.e., off-line handwritten signature verification. Wavelet theory, zero-crossing, dynamic time warping, and nonlinear integer programming form the main body of our methodology. The proposed system can automatically identify useful features which consistently exist within different signatures of the same person and, based on these features, verify whether a signature is a forgery or not. The system starts with a closed-contour tracing algorithm. The curvature data of the traced closed contours are decomposed into multiresolutional signals using wavelet transforms. Then the zero-crossings corresponding to the curvature data are extracted as features for matching. Moreover, a statistical measurement is devised to decide systematically which closed contours and their associated frequency data of a writer are most stable of a writer are most stable and discriminating. Based on these data, the optimal threshold value which controls the accuracy of the feature extraction process is calculated. The proposed approach can be applied to both on-line and off-line signature verification systems.
The second biometric-based pattern recognition problem we deal with is human face recognition; we applied the minimum classification error (MCE) technique proposed by Juang and Katagiri[11]. In this technique, the classical discriminant analysis methodology is blended with the classification rule in a new functional form and is used as the design objective criterion to be optimized by numerical search algorithm. In our work, the MCE formulation is incorporated into a three-layer neural network classifier called multilayer perceptron (MLP). Unlike the traditional probabilistic-based Bayes decision technique, the proposed approach is not necessary to assume the probability model of each class. Besides, the classifier works well even when the size of a training set is small. Moreover, no matter in normal environment or harsh environment, the MCE-based method is superior to the minimum sum-squared error (MSE) based method which is commonly used in traditional neural network classifier. Finally, by incorporating a fast face detection algorithm into the system to help for extracting the face-only image from a complex background, the MCE-based face recognition system is robust to image acquired from harsh environment. Experimental results confirm that our approach outperforms the previous approaches.
| en_US |